Your data, in your own environment.
The controls a security reviewer will ask about, stated plainly: how customer data is separated, encrypted, retained and recovered, who can reach it, and where the boundary sits between your data and any model that learns from it.
Where the boundary sits
Nearly every question asked of a platform that serves many companies reduces to one: is separation enforced by the application, or by the storage underneath it. A filter in application code is one defect away from being no boundary at all.
Each company's data sits in a database of its own. No query can reach from one company's data into another's, because there is nothing shared to reach across.
| Separation | Your data is held in a database of your own, not in a shared one divided up by a column that says which company each row belongs to. |
|---|---|
| Encryption at rest | AES-256. |
| Encryption in transit | TLS 1.2. |
| Backups | Automated and encrypted, every 24 hours. |
| Audit trail | System activity is recorded across the platform. |
Who can reach it
Identity is yours to run. The platform authenticates against your existing directory rather than asking your organization to maintain a second set of credentials, and authorization is expressed in roles you define.
| Onboarding | Self-service account creation and management. |
|---|---|
| Single sign-on | Integrates with your existing organization SSO. |
| Multi-factor authentication | Supported. |
| Authorization | Role-based access control, configurable to your roles. |
What it means for you
If your organization enforces conditional access, device posture or session lifetime at the identity provider, those controls apply here too. We do not ask you to weaken them for us.
What we keep, and what we never do with it
The last row of the table is the one worth reading twice. Your process data, your documents and anything a model derives from them stay inside your own environment. They are not pooled, not used to improve a shared model, and not reachable by another customer.
| Data minimization | We store the process-related data the platform needs and no more. |
|---|---|
| Retention | Configurable retention policies. |
| Residency | Data residency and sovereignty according to the standards of the cloud region your environment runs in. |
| Model boundary | Models fitted on your data stay inside your own environment and are not reachable by any other company. |
This is the difference between an AI feature bolted onto a consumer model and one designed for a regulated industry. A platform that improves for everyone by learning from everyone is, by construction, a platform where your process knowledge leaves your boundary. The architecture here does not permit it.
Staying up, and coming back
Availability commitments matter more than usual when a system holds the record a batch is released against. Replication and failover cover the application and the data together, and recovery is exercised on a schedule rather than assumed.
| Uptime | 99.9% SLA. |
|---|---|
| Replication | Cross-region, for application and data. |
| Failover | Automatic, for application and data. |
| Monitoring | Continuous and automated, with alerting. |
| Recovery testing | Disaster recovery drills every month. |
| Disruption | Proactive mitigation against service disruption. |
Why this shape, for a GxP system
Security controls are necessary but not sufficient for a platform that holds regulated records and puts AI near them. Four architectural properties do the rest of the work, and they are the ones worth testing in diligence.
The record is deterministic
What the platform stores about a run, a method or a batch is what was entered or measured. The AI reads that record and writes alongside it; it does not quietly become the record.
Actions are attributable
Activity is captured in an audit trail, which is the substance of what 21 CFR Part 11 and EU Annex 11 ask of a computerized system holding regulated records.
A person signs
The platform assembles, computes and proposes. Confirmation and signature stay with a named, qualified human, which is also where both FDA and EMA have put the accountable decision.
Answers carry their evidence
Every assertion the AI makes is traceable to the tool result and source record it came from, so a reviewer can check it rather than take it.
Bring us the work you
already know the answer to.
A closed investigation. A released batch. A report your team wrote last quarter. We will run it on the platform, and you judge the output against work you can already check.
or email sales@bioprocess.ai